Explorations of book cover analysis, e.g. detecting when a book has a useful cover image.

cover-analyzer-edges.py

Improved cover analyzer provided by community member. Use instead of cover-analyzer.py.

Requirements

• Python 3
• OpenCV

Example

  cat cover_sources/y2k-useful-covers-100.txt | python cover-analyzer-edges.py > output/results-y2k-useful-covers-100.html

cover-analyzer.py

Takes a list of Internet Archive (Archive.org) item identifiers via stdin, outputs an HTML page containing a table that shows cover images and calculated "usefulness". Some books have colorful covers; others were scanned with just their bare cloth covers. A cover is determined to be "not useful" if it shows too little color variation in its main section.

Requirements

• Python 3
• ImageMagick
• wand (pip install wand)

Example

  cat cover_sources/y2k-useful-covers-100.txt | python cover-analyzer.py > output/results-y2k-useful-covers-100.html

Details

Images undergo the following process:

1. Chop off the top, bottom, and sides, because that's where library stickers usually go.
2. Shrink the remaining image to 16 by 16 pixels, to average out textural differences.
3. Take a standard deviation of the red, green, and blue channels, across all pixels.
4. If all of the standard deviations are below a threshold value, then designate it as a not useful cloth cover.

The cover_sources directory contains files of sample image identifiers:

• not-useful-covers.txt: Books from 1923 with mostly useless covers
• useful-covers.txt: Books from 1923 with a mix of useful and useless covers
• y2k-useful-covers.txt: Modern books with mostly useful covers

The files containing "100" in their filenames are the first 100 identifiers of the corresponding source files. The results of processing these files are in the output directory.

The sample_images directory contains a few sample book cover images.

Contact

Brenton Cheng (brenton at archive dot org)